A nondestructive recognition and classification method for detecting surface defects of Si3N4 bearing balls based on an optimized convolutional neural network

Liao, Dahai; Cui, Zhenzhong; Zhu, Zuoxiang; Jiang, Zhuting; Zheng, Qi; Wu, Nanxing

doi:10.1016/j.optmat.2022.113401

Cited by 3 publications

(2 citation statements)

References 21 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Some scholars have proposed different methods for object-surface-defect detection using supervised deep learning methods. For example, Dahai Liao et al [9] proposed a non-destructive identification and classification method for surface defects based on the improved YOLOv5 algorithm, and they applied a new mobile network attention mechanism, coordinate attention, to the backbone of the YOLOv5 algorithm, giving better detection results. Kechen Song et al [10] proposed a cross-layer semantic guidance network (CS-GNet) based on the YOLOv6 algorithm, which introduces a cross-layer semantic guidance module (CSGM) that uses deeper semantic information to guide the shallower feature layer and improves performance for detecting tiny defects.…”

Section: Related Work 21 Deep Learning Detection Methodsmentioning

confidence: 99%

A Feature-Oriented Reconstruction Method for Surface-Defect Detection on Aluminum Profiles

Tang,

Zhang,

Jin

et al. 2023

Applied Sciences

View full text Add to dashboard Cite

The number of defect samples on the surface of aluminum profiles is small, and the distribution of abnormal visual features is dispersed, such that the existing supervised detection methods cannot effectively detect undefined defects. At the same time, the normal texture of the aluminum profile surface presents non-uniform and non-periodic features, and this irregular distribution makes it difficult for classical reconstruction networks to accurately reconstruct the normal features, resulting in low performance of related unsupervised detection methods. Aiming at such problems, a feature-oriented reconstruction method of unsupervised surface-defect detection method for aluminum profiles is proposed. The aluminum profile image preprocessing stage uses techniques such as boundary extraction, background removal, and data normalization to process the original image and extract the image of the main part of the aluminum profile, which reduces the influence of irrelevant data features on the algorithm. The essential features learning stage precedes the feature-optimization module to eliminate the texture interference of the irregular distribution of the aluminum profile surface, and image blocks of the area images are reconstructed one by one to extract the features through the mask. The defect-detection stage compares the structural similarity of the feature images before and after the reconstruction, and comprehensively determines the detection results. The experimental results improve detection precision by 1.4% and the F1 value by 1.2% over the existing unsupervised methods, proving the effectiveness and superiority of the proposed method.

show abstract

Section: Related Work 21 Deep Learning Detection Methodsmentioning

confidence: 99%

A Feature-Oriented Reconstruction Method for Surface-Defect Detection on Aluminum Profiles

Tang,

Zhang,

Jin

et al. 2023

Applied Sciences

View full text Add to dashboard Cite

show abstract

“…As Si 3 N 4 balls are spherical, manual observation of these defects is difficult. Thus, identifying and classifying these surface defects are complicated, making it difficult to effectively detect each defect type [17].…”

Section: Defectsmentioning

confidence: 99%

Defects and Mechanical Properties of Silicon Nitride Ball Bearings for Electric Vehicle Reducers

Jang,

Lee,

Cheong

et al. 2024

WEVJ

View full text Add to dashboard Cite

In this study, two types of Si3N4 ball bearings for integrated-type EV reducers developed via different manufacturing processes were analyzed to study the microstructure of the balls and the defects that may occur during the manufacturing process. Three types of defects were confirmed that can reduce the fatigue life of Si3N4 ball bearings in operating environments. The microstructure was analyzed to identify the main additive components of each bearing, and pore defects with a size of less than 1 μm, and the types (particle defects or surface defects) and sizes of defects, were analyzed using LSCM, OM, and SEM. Hardness and fracture toughness, which are representative mechanical properties of ceramic materials, were evaluated. The results, evaluated using a Vickers indentation crack-based method, were statistically analyzed to confirm differences in hardness and fracture toughness between the two samples.

show abstract

Research on Image Classification and Recognition Technology Based on Machine Learning

Wang

2024

Applied Mathematics and Nonlinear Sciences

View full text Add to dashboard Cite

In the burgeoning field of computer vision, image classification and recognition have emerged as pivotal areas of research within computer science. This paper introduces a novel approach by amalgamating the Gramian Angular Field (GAF) method, recursive map technique, and Short-Time Fourier Transform (STFT) to devise a joint coding method for null spectrum analysis aimed at extracting spatial features from images. Subsequently, a hybrid classifier combining Deep Convolutional Neural Network (DCNN) and Support Vector Machine (SVM) algorithms—termed 3DCNN-SVM—is constructed to classify and recognize these extracted features. Further, this study conducts a comparative analysis between the proposed feature extraction and classification method and traditional algorithms. The focus of this evaluation is on the domain of vehicle label image recognition, aiming to ascertain the efficacy of the newly developed algorithm. This comprehensive comparison underscores the enhancements in accuracy and efficiency provided by the integration of 3DCNN-SVM in recognizing complex image features, thereby contributing significantly to the advancements in computer vision technology. The results show that after 35 iterations of training, the accuracy of 3DCNN-SVM reaches 99.5% with a loss value of 0.0877. Compared with ShffleNetV2 0.5×, ShffleNetV2 0.5×+ECA, ShffleNetV2 0.5×+ECA +Mish, ShffleNetV2 0.5×+ECA + Mish+SEP are 9.46%, 8.83%, 6.60%, and 5.52% higher on the average accuracy rate of 18 kinds of vehicle label images recognition, respectively. The 3DCNN-SVM classifier model constructed based on machine learning-related theory in this paper realizes the optimization of the traditional classification model and makes the classification and recognition of images more accurate, which has important practical significance.

show abstract

A nondestructive recognition and classification method for detecting surface defects of Si3N4 bearing balls based on an optimized convolutional neural network

Cited by 3 publications

References 21 publications

A Feature-Oriented Reconstruction Method for Surface-Defect Detection on Aluminum Profiles

A Feature-Oriented Reconstruction Method for Surface-Defect Detection on Aluminum Profiles

Defects and Mechanical Properties of Silicon Nitride Ball Bearings for Electric Vehicle Reducers

Research on Image Classification and Recognition Technology Based on Machine Learning

Contact Info

Product

Resources

About